As a fallback for all of the other content extraction techniques, try using the Readability algorithm to extract article text from the HTML page itself.

Video tends to be shared by embedding rather than linking, but it's just as annoying to see the same video more than once.

It would be nice to detect embed code for popular services like YouTube, Vimeo, Slideshare, etc at the start of blog posts and turn those into their own objects which have responses attached, such that all of the posts sharing the same video cluster together in the object stream.

Matt suggested, “There are a couple posts in my feed that are very long, which makes it tough to scan. It'd be nice if it just showed the first 200 words (and whatever media come before then) and hide the rest in a div that I can make magically appear.”

Let's support importing YouTube videos posted and favorited by folks. They have an API so it should be pretty possible.

Some accounts now have invalid auth tokens. This can happen naturally if someone, say, signs in with Twitter then later deletes the Leapfrog app from the apps page inside Twitter (rather than detaching the account from inside Leapfrog).

We have to tolerate transient errors where the token mistakenly didn't work, though. Perhaps we can save the last time the account was successfully polled, and if it was more than a few days ago, stop?

Sometimes there are spurious “-->” sequences in posts. Why is that? Is the HTML parse we're doing to clean up weirding on some comments or something?

As Leapfrog supporting a new content provider (such as youtube, vimeo, flickr), I'd like to see a filtered out content stream. For example, leapf.org/photos will display only photos from twitter (twitpic?), typepad photo assets and flickr photos.

Sometimes I see Instagram photos twice in my stream, as someone will share it both to Flickr (where it's posted as a photo) and to Twitter (where it's posted as a link to the Instagram photo). Since more people share Instagram photos to Twitter than Flickr, perhaps these duplicates could be removed by ignoring Instagram photos on Flickr. (We can tell from the "uploaded:by=instagram" tag.)

It seems that people using non-Twitter services that support OStatus still cross-post their stuff to Twitter because that's the only way to get anyone to read it.

I wonder if there's some way we can recognize when a link is to an OStatus-supporting node and find some way to extract the real content and make that the focus object.

(It's annoying to see a truncated tweet linking to real content on status.net. For example, http://twitter.com/#!/evanpro/status/20986602100625411 .)

Meetup has (undiscoverable) OEmbed, so we should probably add them in, in case people link to meetups.

On my Palm Pre more content won't load even once I scroll to the bottom of the page.

I think it's because the scroll behavior doesn't work in its webkit. Let's try making the "Loading..." message be a "Load more" action you can click, so even if the scroll event doesn't fire you can still tap the message to get more to load.

If you're using keyboard navigation but get to the last item before more items load, and the last item is tall enough to push the “Loading...” message below the fold, trying to “next” can seem broken.

Let's have the keyboard navigation scroll to the bottom in that case maybe.

Currently when an item is created from a feed entry that only has a summary, it's created as a “link” item. Some of these posts are actually putting their whole text in the summary. Perhaps, if the summary contains enough HTML, we should make it be a full “mixed” item.

We've added a few feedback actions you can take on items, but to support reblogging something to TypePad we'd have to reverse engineer how to invoke the reblog bookmarklet.

Since that means we have to reblog by URL, it wouldn't really work like the other items anyway, so we could do it for any mixed content item, not only TypePad posts.

Matt suggested, “I'd like it if the background color for the comments was set to .75 opacity so they blend in a bit more.”

Amazon doesn't provide good opengraph data, but can we use the Amazon API to have better data for Amazon link objects?

Currently we can't pull feed content on sites that use Feedburner feeds, because Feedburner always opaques the URLs through their redirector, and we don't want to test every redirector URL to see which one goes to the real article.

If we can figure out some other way to support Feedburner feeds, we should. (If we implement Readability style content finding, this is less of an issue, even though we won't be have “official” data such as it is.)